🤖 Former OpenAI Researcher Admits to Copyright Violations

Subscribe | AI Toolkit | Meet The Team

Welcome back AI enthusiasts!

In today’s AI Report:

• 🎙Former OpenAI Researcher Admits to Copyright Violations

• 🎥Runway Introduces “Act-One” to Generate Expressive Characters

• 🔍Google DeepMind’s AI Watermarking Tool

• 🛠Trending Tools

• 💰Funding Frontlines

• 💼Who’s Hiring?

Read Time: 3 minutes

🗞RECENT NEWS

OPENAI

🎙Former OpenAI Researcher Admits to Copyright Violations

A former OpenAI Researcher who helped collect, filter, and deploy training datasets for ChatGPT has spoken out against the company.

Key Details:

• Suchir Balaji, a 25-year-old who joined OpenAI in 2020 after graduating from the University of California, Berkeley (i.e., “UC Berkeley”), voiced concerns about the company’s business practices.

• He believes that OpenAI’s use of copyrighted content is illegal and ChatGPT is damaging how we consume content across the Internet.

• Earlier this week, Balaji published an essay detailing how copyrighted content in training datasets influences ChatGPT’s outputs and why it’s damaging the content ecosystem.

• “I think it’s pretty obvious that the Market Harms from ChatGPT mostly come from it producing substitutes,” said Balaji. Market Harms are unintended side effects of market interactions.

• “For example, if we had the programming question, ‘Why does 0.1 + 0.2 = 0.30000000000000004 in floating point arithmetic?’, we could ask ChatGPT instead of searching Stack Overflow.”

• He explained that “the Market Harms from this type of use can be measured in decreased website traffic to Stack Overflow.”

• More importantly, ChatGPT relied on training datasets of Stack Overflow’s content to answer this programming question.

🚨Read Balaji’s essay here.

🩺 PULSE CHECK

RUNWAY

🎥Runway Introduces “Act-One” to Generate Expressive Characters

Runway introduced “Act-One,” a new way to place your detailed facial expressions onto AI-generated characters.

Key Details:

• “Act-One” enables anyone to transfer their facial expressions onto AI-generated characters by leveraging their smartphone’s camera.

• It’s able to “translate the performance from a single video across countless different AI-generated character designs in many different styles.”

• “Act-One” directly integrates with Runway’s “Gen-3 Alph,” which converts text prompts or still images into highly detailed videos with complex scenes and cinematic movements.

Why It’s Important:

• Traditionally, facial animations require extensive processes, including manual face rigging, motion capture equipment, and multiple reference footages. But now, anyone can do it!

• Lionsgate recently partnered with Runway to explore the use of AI in film production. This partnership allows Runway to build AI models trained on Lionsgate’s extensive movie library, which includes The Hunger Games Series, Hacksaw Ridge, and the John Wick Saga.

🚨Watch it in action here.

AI RESEARCH

🔍Google DeepMind’s Invisible Watermarking for AI-Generated Text

Google DeepMind developed “SynthID-Text,” which adds invisible watermark labels to AI-generated text by encoding Tokens. Tokens are the smallest units of data used by an AI model to process and generate text. Similarly, we break down sentences into words or characters.

“SynthID-Text” uses a Tournament Sampling approach that embeds undetectable watermarks while preserving the quality of AI-generated text.

Tournament Sampling involves three steps:

1. Options: The AI model generates several possible alternatives for each word in the sentence.

2. Tournament: These alternatives “compete” against each other.

3. Selection: The winning alternative, determined by factors like context and grammar, is chosen as the final word.

In simple terms, the AI model alters which words it chooses to generate when creating sentences. This technique creates unique patterns that can be detected with a Cryptographic Key. A Cryptographic Key is a string of characters used within an encryption framework that alters data to make it appear random.

So, why is this important? “SynthID-Text” provides a way to trace the source of AI-generated content, allowing for the proper attribution and credit.

🛠TRENDING TOOLS

📱StoryShort creates viral faceless videos.

🗣Hedy is an AI assistant that helps you excel in meetings.

🎙Podial turns documents into engaging podcast discussions.

🗺MyPeas transforms your job description into an actionable AI roadmap.

📃PaperGuide discovers, creates, writes, and manages research with ease.

🔮Browse our always Up-To-Date AI Tools Database.

💰FUNDING FRONTLINES

• Carbon Robotics raises a $70M Series D to offer AI-powered farming solutions.

• Alimetry secures an $18M Series A for an AI wearable that diagnoses gastric disorders.

• DataBank lands nearly $2B to create an Investment Fund that finances datacenter expansions.

💼WHO’S HIRING?

• JellyFish (Remote): Data Science Co-Op, Summer 2025

• Matroid (Palo Alto, CA): Computer Vision {CV} Intern, Summer 2025

• IBM (San Jose, CA): Research Scientist, Large Scale Language Models Intern, Summer 2025

• Nvidia (Santa Clara, CA): Math Libraries, Quantum Computing, New College Grad 2025

• Brain Corp. (San Diego, CA): Software Engineer I, Applied Machine Learning {ML}

🤖PROMPT OF THE DAY

EMAIL COMMUNICATION

📬Effective Email Communication

Develop an effective email communication strategy tailored for [Small Business] with [Product/Service] in [Industry] with [Target Audience]. It should include best practices for [Specific Types of Emails], such as tone, content, and frequency.

Small Business = [Insert Here]

Product/Service = [Insert Here]

Industry = [Insert Here]

Target Audience = [Insert Here]

Specific Types of Emails = [Insert Here]

📒FINAL NOTE

FEEDBACK

❤️TAIP Review of The Day

REFER & EARN

🎉Your Friends Learn, You Earn!

You currently have 0 referrals, only 1 away from receiving ⚙️Ultimate Prompt Engineering Guide.

Refer 3 friends to learn how to 👷‍♀️Build Custom Versions of OpenAI’s ChatGPT.

Copy and paste this link to friends: https://theaipulse.beehiiv.com/subscribe?ref=PLACEHOLDER